Last verified: October 4, 2026
DeepSeek V4 is best understood as the DeepSeek model family that introduced DeepSeek-V4-Pro and DeepSeek-V4-Flash in April 2026. But if you are choosing a DeepSeek API model today, “DeepSeek V4” is no longer a single static model name. The current hosted API picture includes deepseek-flash for DeepSeek-V4.1-Flash, deepseek-v4-pro for DeepSeek-V4-Pro-0813, and retired compatibility aliases such as deepseek-v4-flash and deepseek-v4-flash-vision-exp.
The practical answer: for most new Flash-style API work, start with deepseek-flash, because DeepSeek’s current documentation lists it as the model name for DeepSeek-V4.1-Flash. If you specifically want Pro behavior, check the current DeepSeek API Change Log and Models & Pricing page before deploying. DeepSeek’s V4.1 Flash announcement said V4 Pro would be phased out and routed to V4.1 Flash, but the API changelog later says DeepSeek will continue providing API services for deepseek-v4-pro after September 14, 2026, with billing unchanged.
Do not copy old V4 examples blindly. In particular, avoid treating deepseek-v4-flash, deepseek-v4-flash-vision-exp, deepseek-chat, or deepseek-reasoner as current default model names for new V4 code unless the current documentation explicitly supports that use.
Quick answer: is DeepSeek V4 still current?
Yes, but only if you define what you mean by “DeepSeek V4.”
The original DeepSeek-V4 Preview release from April 24, 2026 introduced V4-Pro and V4-Flash. Since then, DeepSeek released V4-Flash updates, V4-Pro GA, V4-Flash-Vision-Exp, and then DeepSeek-V4.1-Flash. That means a current DeepSeek V4 article must explain the model lifecycle, not just repeat the April 2026 launch notes.
For current API usage, the safest rule is:
- Use
deepseek-flashwhen you want the current V4.1 Flash hosted API model. - Check current docs before using
deepseek-v4-pro, because its status changed around the V4.1 Flash rollout. - Do not use
deepseek-v4-flashordeepseek-v4-flash-vision-expas preferred new model names; DeepSeek says those model names are legacy compatibility aliases routed to V4.1 Flash. - Do not use
deepseek-chatordeepseek-reasoneras DeepSeek V4 entry points for new code; DeepSeek’s April 2026 V4 entry described them as legacy names scheduled for retirement.
DeepSeek V4 status table
This table separates the model family, hosted API model IDs, retired aliases, and open-weight model-card references. That distinction matters because hosted API routing and model-card availability are not the same thing.
| Item | What it is | Current official status | Use today? |
|---|---|---|---|
| DeepSeek V4 | The broader V4 model family introduced with V4-Pro and V4-Flash. | Still relevant as a lifecycle term, but not a single current model ID. | Use the term for the family/history, not as an API model name. |
| DeepSeek V4 Preview | April 24, 2026 preview release with V4-Pro and V4-Flash. | Historical release. It introduced the original V4 model family and 1M-context positioning. | Use for background only; do not treat it as the full current state. |
| DeepSeek-V4-Pro | Higher-capability V4 route, later released as GA. | DeepSeek’s August 13, 2026 changelog says V4-Pro GA rolled out on App, Web, and API using deepseek-v4-pro. | Use only after checking the current Models & Pricing table and changelog. |
deepseek-v4-pro | Hosted API model ID for DeepSeek-V4-Pro-0813 in the current pricing table. | Current Models & Pricing lists it as DeepSeek-V4-Pro-0813. The changelog says API service continues after September 14, 2026 with billing unchanged. | Usable if current account/API docs still list it for your account and workload. |
| DeepSeek-V4-Flash | Original Flash route from the V4 family. | Superseded by V4.1 Flash in current hosted API guidance. | Do not use as the preferred current hosted API target. |
deepseek-v4-flash | Old V4 Flash API model name. | DeepSeek says it is still accepted, but the corresponding model is retired and requests are served by V4.1 Flash and billed at Flash price. | Avoid in new code; use deepseek-flash. |
| DeepSeek-V4-Flash-Vision-Exp | Experimental multimodal route announced in August 2026. | Superseded by V4.1 Flash’s native multimodal support in current hosted API guidance. | Do not use as a preferred new model route. |
deepseek-v4-flash-vision-exp | Old experimental vision API model name. | DeepSeek says it is still accepted for compatibility, but the model is retired and requests are served by V4.1 Flash. | Avoid in new code; use deepseek-flash for current Flash/vision usage. |
| DeepSeek-V4.1-Flash | Current V4.1 Flash model with native multimodal support. | Released September 10, 2026. DeepSeek says it improves text and agent performance and supports native multimodal visual understanding. | Usually the safest current Flash default. |
deepseek-flash | Current API model ID for DeepSeek-V4.1-Flash. | Listed in Models & Pricing as DeepSeek-V4.1-Flash. | Use for new Flash-style API integrations. |
| DeepSeek-V4.1-Pro | Future/larger V4.1 Pro route mentioned in V4.1 Flash rollout context. | No official source checked for this article listed V4.1-Pro as an available API model. | Do not write code for it unless the current docs list it. |
deepseek-chat | Old general chat alias. | DeepSeek’s April 2026 V4 entry said it would be discontinued and temporarily pointed to V4-Flash non-thinking mode during the transition. | Do not use as a V4 entry point for new code. |
deepseek-reasoner | Old reasoning alias associated with earlier R1/V3-era usage. | DeepSeek’s April 2026 V4 entry said it would be discontinued and temporarily pointed to V4-Flash thinking mode during the transition. | Do not use as a current V4 entry point unless current docs explicitly say so. |
What DeepSeek V4 originally introduced
The original DeepSeek-V4 Preview release introduced two main routes: DeepSeek-V4-Pro and DeepSeek-V4-Flash. DeepSeek described V4-Pro as a 1.6T-total-parameter model with 49B active parameters, and V4-Flash as a 284B-total-parameter model with 13B active parameters. The same launch positioned V4 as a cost-effective 1M-context family and said the API was available by setting the model to deepseek-v4-pro or deepseek-v4-flash. DeepSeek-V4 Preview release note
That April release also said the base URL stayed the same and that V4 supported both OpenAI ChatCompletions and Anthropic API formats. It also warned that deepseek-chat and deepseek-reasoner would be retired after a transition period, with those older names temporarily routing to V4-Flash non-thinking and thinking modes.
The important point: the April 2026 V4 Preview was a starting point, not the final state of the V4 line.
What changed after the original V4 release?
DeepSeek updated the V4 family several times after the preview release. These updates changed which model names developers should use and how they should interpret older articles.
| Date | Update | Why it matters |
|---|---|---|
| April 24, 2026 | DeepSeek-V4 Preview launched. | Introduced V4-Pro and V4-Flash, with API model names deepseek-v4-pro and deepseek-v4-flash. |
| July 31, 2026 | DeepSeek-V4-Flash API public beta update. | The changelog says V4-Flash kept the same architecture and size as the preview but was re-post-trained, and added native Responses API support. |
| August 13, 2026 | DeepSeek-V4-Pro GA. | The changelog says V4-Pro rolled out on App, Web, and API, and that users could call it with deepseek-v4-pro. |
| August 21, 2026 | DeepSeek-V4-Flash-Vision-Exp released. | Added an experimental multimodal vision route using deepseek-v4-flash-vision-exp. |
| September 10, 2026 | DeepSeek-V4.1-Flash released. | Made deepseek-flash the current V4.1 Flash API model name and retired old V4 Flash aliases for compatibility routing. |
The September 10 update is the most important for current users. DeepSeek says V4.1 Flash is live on the API with native multimodal support, and tells users to set the model to deepseek-flash. It also says the previous-generation V4 Flash and V4 Flash Vision Exp models are retired, while deepseek-v4-flash and deepseek-v4-flash-vision-exp are temporarily routed to V4.1 Flash for compatibility. DeepSeek V4.1 Flash release note
Current API model names: deepseek-flash, deepseek-v4-pro, and retired aliases
For new API integrations, the current model-name guidance is clear for Flash and more nuanced for Pro.
deepseek-flash is the current model ID for DeepSeek-V4.1-Flash. The Models & Pricing page lists deepseek-flash as DeepSeek-V4.1-Flash and says legacy names such as deepseek-v4-flash and deepseek-v4-flash-vision-exp are still accepted, but their underlying models have been retired and the requests are served by V4.1 Flash at Flash pricing. DeepSeek Models & Pricing
deepseek-v4-pro is more complicated. The V4.1 Flash release note said V4 Pro was being phased out and that deepseek-v4-pro requests would route to V4.1 Flash starting September 14, 2026 until V4.1-Pro launches. But the API Change Log currently says DeepSeek decided to continue providing API services for DeepSeek V4 Pro after September 14, 2026, with the billing method unchanged. That is why you should check the current Models & Pricing table before deploying code that depends on Pro behavior. DeepSeek API Change Log
In practice, configure model IDs centrally. Do not hard-code model names throughout your codebase. That makes it easier to switch from retired aliases to deepseek-flash, or to confirm deepseek-v4-pro behavior if your production workload needs it.
DeepSeek V4 Pro vs V4 Flash vs V4.1 Flash
The names are similar, but they point to different moments in the model lifecycle.
| Model / route | Best understood as | API guidance today |
|---|---|---|
| DeepSeek-V4-Pro Preview | The April 2026 preview Pro model. | Historical context only. Do not use preview-era assumptions for current production decisions. |
| DeepSeek-V4-Pro-0813 | The August 2026 GA Pro model. | Current Models & Pricing lists deepseek-v4-pro as DeepSeek-V4-Pro-0813. Verify current routing before use. |
| DeepSeek-V4-Flash Preview | The original lower-cost V4 Flash route. | Historical context. Superseded by later Flash updates and V4.1 Flash. |
| DeepSeek-V4-Flash-0731 | Public beta Flash update from July 2026. | Superseded in current hosted API guidance by V4.1 Flash. |
| DeepSeek-V4-Flash-Vision-Exp | Experimental vision route from August 2026. | Retired as a separate current model route; compatibility alias routes to V4.1 Flash. |
| DeepSeek-V4.1-Flash | September 2026 Flash model with native multimodal support. | Use deepseek-flash for new Flash API work. |
For most developers, the decision is not “V4 Pro or V4 Flash?” anymore. The better question is: should this workload use current deepseek-flash, current deepseek-v4-pro, or a local/open-weight deployment?
Feature comparison: context, output, tools, vision, and API formats
DeepSeek’s current Models & Pricing page lists the practical feature differences between deepseek-flash and deepseek-v4-pro.
| Feature | deepseek-flash | deepseek-v4-pro |
|---|---|---|
| Model version | DeepSeek-V4.1-Flash | DeepSeek-V4-Pro-0813 |
| OpenAI-format base URL | https://api.deepseek.com | https://api.deepseek.com |
| Anthropic-format base URL | https://api.deepseek.com/anthropic | https://api.deepseek.com/anthropic |
| Thinking mode | Supports thinking and non-thinking modes; thinking is default according to the pricing table. | Supports thinking and non-thinking modes; thinking is default according to the pricing table. |
| Context length | 1M | 1M |
| Maximum output | 384K | 384K |
| JSON Output | Supported | Supported |
| Tool Calls | Supported | Supported |
| Responses API | Supported | Supported |
| Anthropic API | Supported | Supported |
| Chat Prefix Completion | Supported | Supported |
| FIM Completion | Non-thinking mode only | Non-thinking mode only |
| Vision | Supported | Not supported in the current pricing table |
| Concurrency limit | 2500 | 500 |
The biggest current feature difference is vision: deepseek-flash supports it, while deepseek-v4-pro is listed as not supporting vision. For multimodal workflows, the current default should normally be deepseek-flash.
Pricing and cost considerations
DeepSeek currently prices API usage per 1M tokens and separates input cache-hit tokens, input cache-miss tokens, and output tokens. It also uses peak and off-peak pricing.
| Model ID | Input cache hit | Input cache miss | Output | Practical note |
|---|---|---|---|---|
deepseek-flash off-peak | $0.003 | $0.15 | $0.60 | Lowest current hosted DeepSeek V4-family route in the official table. |
deepseek-flash peak | $0.006 | $0.30 | $1.20 | Peak pricing doubles off-peak pricing. |
deepseek-v4-pro off-peak | $0.022 | $0.66 | $1.98 | Higher-cost Pro route listed as DeepSeek-V4-Pro-0813. |
deepseek-v4-pro peak | $0.044 | $1.32 | $3.96 | Peak pricing doubles off-peak pricing. |
DeepSeek says peak hours are 01:00–04:00 and 06:00–10:00 UTC, Monday through Friday, excluding Chinese public holidays. All other hours are off-peak. Pricing can change, so production teams should regularly check the official pricing page rather than relying on old V4 launch screenshots.
For cost-sensitive workloads, deepseek-flash is usually the better first choice. For tasks where you believe Pro behavior materially improves success rate, evaluate deepseek-v4-pro on your own prompts and compare total cost per successful task, not just per-token price.
Open weights and local deployment
Do not confuse hosted API model IDs with open-weight checkpoints. A hosted API route can be retired, redirected, or priced differently while a model card remains available for research or local deployment.
The DeepSeek-V4.1-Flash model card on Hugging Face lists an MIT license and shows usage paths for libraries, inference providers, notebooks, local apps, vLLM, Docker Model Runner, and SGLang. The V4.1 Flash release note describes it as a 552B-parameter MoE model with an asymmetric architecture using 8B active parameters for input and 16B for output.
The DeepSeek-V4-Pro-0813 model card is also live on Hugging Face and lists an MIT license. It includes examples for Transformers, vLLM, Docker Model Runner, and SGLang.
Open weights do not mean easy or free deployment. Large models can require serious GPU, storage, serving, monitoring, and security work. For most teams, the hosted API is the practical starting point; open-weight deployment is mainly for teams with a clear infrastructure reason to run their own stack.
When should you use deepseek-flash?
Use deepseek-flash as the default starting point when you want current DeepSeek hosted API access and do not have a proven reason to pay for Pro. It is the current V4.1 Flash model ID, supports vision in the current model table, has higher documented concurrency than Pro, and has much lower token prices.
deepseek-flash is usually the best first choice for:
- Cost-sensitive chat applications.
- High-throughput API workloads.
- Agent workflows where many calls repeat context.
- Vision or image-and-text input workflows.
- Long-context tasks where input cost matters.
- OpenAI-compatible integrations that should follow current model naming.
- New projects that do not need Pro-specific evaluation.
When should you use deepseek-v4-pro?
Use deepseek-v4-pro only when your workload needs the current Pro route and you have verified that the current DeepSeek docs still support it for your account and API path. The current pricing table lists deepseek-v4-pro as DeepSeek-V4-Pro-0813, but the V4.1 Flash rollout introduced enough status confusion that new production deployments should verify the changelog before assuming behavior.
deepseek-v4-pro may be worth testing for:
- Hard coding-agent tasks.
- Scientific or technical reasoning where your tests show Pro wins.
- Complex agent workflows where answer quality matters more than token cost.
- Existing applications already tuned around Pro outputs.
- Benchmarks or evaluations that specifically require V4-Pro-0813.
Do not assume Pro is better for every workload. DeepSeek’s own V4.1 Flash release says V4.1 Flash was designed for higher capability, faster inference, higher throughput, and lower cost. The right choice should come from your own evaluation set.
What old V4 tutorials get wrong
Many older DeepSeek V4 tutorials are now incomplete because they were written before the V4.1 Flash rollout or before the V4-Pro GA update.
| Old advice | Why it may be wrong now | What to do instead |
|---|---|---|
Use deepseek-v4-flash for Flash. | DeepSeek says the old V4 Flash model is retired and the alias routes to V4.1 Flash. | Use deepseek-flash in new code. |
Use deepseek-v4-flash-vision-exp for vision. | The experimental vision route has been retired as a separate current model; V4.1 Flash has native multimodal support. | Use deepseek-flash and the current Vision/Responses API docs. |
Use deepseek-chat or deepseek-reasoner for V4. | DeepSeek’s April 2026 V4 entry described those as legacy names scheduled for retirement. | Use current model IDs from Models & Pricing. |
| Assume V4 Pro was fully removed after V4.1 Flash. | The V4.1 Flash blog and API changelog use different wording. The changelog says V4 Pro API service continues after September 14, 2026. | Check the current changelog and pricing table before deployment. |
| Use old V4 Preview pricing. | DeepSeek changed pricing after V4-Pro GA and again after V4.1 Flash. | Use the current Models & Pricing page. |
| Call V4 “open source” without qualification. | Open weights, hosted API access, license terms, and local deployment are different things. | Say “open weights” when referring to model-card checkpoints. |
Which DeepSeek model should you use today?
Use the model that matches the job, not the most famous name.
| Your need | Best starting point | Why |
|---|---|---|
| New hosted API project | deepseek-flash | Current model ID for V4.1 Flash, lower cost, vision support, high concurrency. |
| Multimodal / image input | deepseek-flash | Current model table lists Vision support for Flash, not Pro. |
| Cost-sensitive long-context workload | deepseek-flash | Lower cache-hit, cache-miss, and output pricing. |
| Hard agentic coding or reasoning test | Test deepseek-v4-pro and deepseek-flash | Pro may help some workloads, but you should verify with your evaluation set. |
| Existing Pro-tuned application | deepseek-v4-pro, after verification | The current pricing table lists it, but status/routing should be checked before production changes. |
| Local or open-weight experimentation | V4.1 Flash or V4-Pro model-card checkpoints | Use official model cards, license terms, and infrastructure planning. |
| Old R1-style reasoning examples | Current model + Thinking Mode | Do not use old aliases without checking current docs. |
Practical migration checklist
- Search your codebase for
deepseek-v4-flash,deepseek-v4-flash-vision-exp,deepseek-chat, anddeepseek-reasoner. - Replace old Flash routes with
deepseek-flashunless you have a documented compatibility reason not to. - Keep
deepseek-v4-proonly if the current changelog and Models & Pricing page support your use case. - Move model names into configuration so you can change them without editing application logic.
- Re-run quality, latency, tool-call, and cost evaluations after model routing changes.
- Re-check pricing before running large production workloads.
- Use the current Thinking Mode guide instead of old
deepseek-reasonerexamples. - Do not assume open-weight model-card behavior is identical to hosted API behavior.
- For sensitive or business data, review the current DeepSeek privacy and platform terms before use.
Frequently asked questions
What is DeepSeek V4?
DeepSeek V4 is the model family DeepSeek introduced in April 2026 with DeepSeek-V4-Pro and DeepSeek-V4-Flash. Today, the term also requires lifecycle context because DeepSeek later released V4-Pro GA, V4-Flash updates, V4-Flash-Vision-Exp, and DeepSeek-V4.1-Flash.
Is DeepSeek V4 still available?
Parts of the V4 family remain relevant, but you should use current model IDs. DeepSeek’s current pricing table lists deepseek-flash for DeepSeek-V4.1-Flash and deepseek-v4-pro for DeepSeek-V4-Pro-0813.
Is DeepSeek V4 the same as DeepSeek V4.1 Flash?
No. DeepSeek V4 refers to the broader V4 family that began with V4-Pro and V4-Flash. DeepSeek-V4.1-Flash is a later Flash model in the V4.1 generation, called through the current API model name deepseek-flash.
What is DeepSeek V4 Pro?
DeepSeek V4 Pro is the higher-capability V4 route. The current Models & Pricing page lists deepseek-v4-pro as DeepSeek-V4-Pro-0813. The August 2026 changelog says V4-Pro GA rolled out on App, Web, and API.
Is deepseek-v4-pro still working?
The current official Models & Pricing page lists deepseek-v4-pro, and the API changelog says DeepSeek decided to continue providing API services for DeepSeek V4 Pro after September 14, 2026. Because V4 Pro status changed during the V4.1 Flash rollout, verify the current changelog before production deployment.
What is DeepSeek V4 Flash?
DeepSeek V4 Flash was the lower-cost Flash route in the original V4 family. It has since been superseded in hosted API guidance by DeepSeek-V4.1-Flash, called with deepseek-flash.
Is deepseek-v4-flash retired?
Yes, as a current model route. DeepSeek says deepseek-v4-flash is still accepted for compatibility, but the old V4 Flash model is retired and requests are served by DeepSeek-V4.1-Flash and billed at Flash price.
What is deepseek-flash?
deepseek-flash is the current hosted API model name for DeepSeek-V4.1-Flash. It is the recommended current Flash model ID in DeepSeek’s documentation.
Should I use deepseek-flash or deepseek-v4-pro?
Use deepseek-flash first for most new API projects, especially cost-sensitive, high-throughput, long-context, or vision workflows. Test deepseek-v4-pro if you specifically need Pro behavior and current docs confirm it fits your use case.
What happened to deepseek-chat and deepseek-reasoner?
DeepSeek’s April 2026 V4 entry said deepseek-chat and deepseek-reasoner were legacy model names scheduled for retirement, temporarily routing to V4-Flash non-thinking and thinking modes during the transition. Do not use them as current V4 entry points for new code unless current docs explicitly say so.
Does DeepSeek V4 support 1M context?
Yes. The original V4 release described 1M context as the standard across official DeepSeek services, and the current Models & Pricing table lists 1M context length for deepseek-flash and deepseek-v4-pro.
Does DeepSeek V4 support Thinking Mode?
Yes. The current Models & Pricing page says both deepseek-flash and deepseek-v4-pro support thinking and non-thinking modes. The Thinking Mode guide explains how to control effort levels and switch modes.
Does DeepSeek V4 support the Responses API?
Yes. DeepSeek’s current Models & Pricing page lists Responses API support for both deepseek-flash and deepseek-v4-pro. The V4-Flash and V4-Pro changelog entries also describe native Responses API support and Codex adaptation.
Does DeepSeek V4 support Anthropic API format?
Yes. The current pricing table lists an Anthropic-format base URL, https://api.deepseek.com/anthropic, for the current listed models.
Does DeepSeek V4 have open weights?
DeepSeek has official model cards for DeepSeek-V4.1-Flash and DeepSeek-V4-Pro-0813 on Hugging Face, and those model cards list MIT licenses. Open weights are different from hosted API access and still require infrastructure if you self-host.
Can I run DeepSeek V4 locally?
Official Hugging Face model cards provide local and serving paths such as Transformers, vLLM, Docker Model Runner, and SGLang for V4.1 Flash and V4-Pro-0813. That does not mean local deployment is easy or cheap; large models require serious infrastructure planning.
Is DeepSeek V4 good for coding agents?
DeepSeek’s V4 changelog repeatedly emphasizes agent and coding-agent improvements, especially in V4-Flash, V4-Pro, and V4.1 Flash. Treat those as DeepSeek-reported claims and test your own coding-agent workflow before choosing a production model.
Is V4.1-Pro available?
No official source checked for this article listed DeepSeek-V4.1-Pro as an available API model. The V4.1 Flash release note mentions V4.1-Pro as a future point, but current model choices should come from the Models & Pricing page and API changelog.
Bottom line
DeepSeek V4 is still important, but it is not one fixed model name anymore. It is a model family and lifecycle that began with V4-Pro and V4-Flash, then moved through Flash updates, V4-Pro GA, V4-Flash-Vision-Exp, and V4.1 Flash.
For new hosted API work, start with deepseek-flash unless your workload has a specific reason to use Pro. For deepseek-v4-pro, verify the current changelog and Models & Pricing page before deployment. For old aliases such as deepseek-v4-flash, deepseek-v4-flash-vision-exp, deepseek-chat, and deepseek-reasoner, treat old tutorials as migration material, not current model-selection guidance.